Nebius Raised Prices Again - Now $7.8 Billion Has To Earn Its Keep

Nebius will raise on-demand Nvidia GPU prices again in October, signaling sustained AI-cloud demand as capacity expands. Q2 AI cloud revenue was $574.9M and adjusted EBITDA margin approached 50%, while $7.8B of infrastructure remained not yet in use. Management is preserving some uncontracted future capacity to capture potentially higher short-term GPU pricing.

Analysis

NBIS is shifting from a contracted-infrastructure model toward a higher-beta merchant GPU model. That should expand realized revenue per GPU and accelerate payback on deployed clusters if spot demand remains tight, but it also makes quarterly results materially more sensitive to utilization: a 5-10 point utilization shortfall can erase much of the apparent incremental pricing benefit because depreciation, power commitments and network costs are largely fixed. The relevant valuation question is therefore not the announced price increase, but whether revenue per installed GPU rises faster than the company’s capital base and depreciation expense over the next two quarters.

The competitive read-through is favorable for NVDA’s accelerator demand but only modestly so: pricing power at a smaller cloud provider confirms scarcity at the inference/training capacity layer, yet does not by itself alter NVDA’s supply outlook. More directly, it challenges CRWV and other GPU-cloud operators to defend pricing while funding rapid capex; NBIS’s willingness to retain uncontracted capacity could capture upside in a shortage, but it also introduces price competition if industry supply arrives ahead of enterprise workloads. Hyperscalers such as MSFT, AMZN and GOOGL are insulated by broader cloud bundles and balance sheets, while merchant providers bear the residual-demand risk.

Near term, October implementation and the next utilization/revenue-per-GPU disclosure are catalysts for NBIS. Over 1-3 months, the stock can rerate if management demonstrates that incremental capacity is absorbed without a decline in adjusted EBITDA margin; over 6-18 months, the principal risk is that a large depreciating asset base requires sustained premium pricing to avoid FCF disappointment. The contrarian view is that investors may be underwriting scarcity as structural when new GPU capacity, customer-owned clusters, or lower-cost inference silicon could normalize merchant pricing faster than consensus expects.

Thesis falsification: reduce the constructive view if NBIS reports sequential revenue per GPU or utilization deterioration despite the price action, materially lowers margin/FCF expectations, or signals that available capacity must be contracted at discounted rates. A sustained compression in CRWV pricing or accelerating hyperscaler GPU-capex supply would also weaken the scarcity premise.

Key Decisions for Investors

• Initiate a tactical long NBIS only after the October pricing change is visible in customer uptake or management confirms stable-to-higher utilization; target a 1-3 month catalyst window through the next earnings update. Size modestly given the merchant-capacity exposure; exit if revenue per GPU or adjusted EBITDA margin declines sequentially, rather than relying on aggregate revenue growth.
• Express the cleaner industry confirmation via long NVDA versus short CRWV over 3-6 months if GPU-cloud pricing remains firm. NVDA monetizes broad accelerator demand with less utilization and financing risk, while CRWV carries greater sensitivity to funding costs, contract concentration and merchant-cloud price competition; cover the short if CRWV demonstrates sustained utilization gains with improving FCF conversion.
• Do not chase NBIS solely on a published price increase. Set an alert for disclosed utilization, contracted backlog, and capex/depreciation guidance: a price increase accompanied by expanding idle capacity is a negative operating-leverage signal, not a bullish demand signal.
• For existing NBIS longs, consider trimming into a sharp pre-earnings rerating and retaining upside through defined-risk calls only if liquidity permits. The asymmetric downside is a multiple reset from a utilization miss, whereas the upside case requires independently verifiable evidence that premium pricing persists across newly activated capacity.